HJ-Xu/WTFM-Layer

WTFM-Layer

Code for "WTFM Layer: An Effective Map Extractor for Unsupervised Shape Correspondence", published at PG 2022.

You can view the detailed content of the paper here.

Installation

python  >= 3.8
cuda    =  10.2 
pytorch >= 1.10.0

Download data

Regarding the remesh 5K dataset, we used the dataset from GeomFmaps, and for the anisotropic dataset, we used the dataset from DUO-FMNet. Please put the downloaded data in the off format into the directory data/datasetname/shapes.

Usage

To train WTFM model, use the training script:

> python train.py  

We provide networks trained on the Faust Remesh and SCAPE Remesh datasets, and the network parameters are saved in saved_models/faust/ckpt_ep0.pth and saved_models/scape/ckpt_ep0.pth. You can directly run the following code to verify the experimental accuracy mentioned in the paper:

> python train.py --evaluate

Citation

@inproceedings{liu2022wtfm,
  title={WTFM Layer: An Effective Map Extractor for Unsupervised Shape Correspondence},
  author={Liu, Shengjun and Xu, Haojun and Yan, Dong-Ming and Hu, Ling and Liu, Xinru and Li, Qinsong},
  booktitle={Computer Graphics Forum},
  volume={41},
  number={7},
  pages={51--61},
  year={2022},
  organization={Wiley Online Library}
}